Geography Reference
In-Depth Information
Table 3.9
The attribute Gain Ratio value for constructing decision tree
Name
Gain ratio
Rank
Description
X
0.03
8
Rectangular coordination of Longitude
Y
0.22
2
Rectangular coordination of latitude
PA
0.12
6
0.1mmannual precipitation
TA
0.20
3
0.1 C annual accumulated temperature
DEM
0.15
4
Elevation
LFM
0.11
7
Landform type
NDVI-3
0.22
2
Normalized differential vegetation index in March
NDVI-12
0.27
1
Normalized differential vegetation index in December
Net primary productivity (g C/m
2
/year)
NPP
0.14
5
Gain
ð
A
Þ
¼I
ð
S
1
;
S
2
;
...
;
S
m
Þ
E
ð
A
Þ
ð
3
:
16
Þ
The attribute A contains the DEM, longitude, latitude, annual temperature,
annual precipitation, NPP, NDVI and other ancillary spatial data. We calculate the
gain ratio to select the attributes that can generate the ancillary information of
classification (Table
3.9
). There are about 35,396 sample cells of the closed forest
and other forest. The gain ratio for training dataset is calculated, the biggest value
of which is 0.27, indicating that NDVI-12 is the most suitable to be the attribute
for the forest categories. The forest is further divided into two sub-categories
according to the NDVI-12 and NDVI-3, i.e., the forest with the NDVI-12 reaching
0.53 and NDVI-3 reaching 0.39 is categorize into the evergreen forest, while the
forest with the NDVI-12 below 0.53 and NDVI-3 reaching below is categorized
into the deciduous forest. Although the gain ratio of DEM and temperature is
higher than that of the NPP, it is difficult to distinguish the forest type according to
them. Therefore, we distinguish Broadleaved, Needleleaved and mixed forest
according to the NPP. Broadleaved forest is more than 445, and that of Needle-
leaved forest is less than 297, and the forests with the middle NPP value is
categorized into the mixed forest.
The accuracy of different classifiers is compared with the WEKA toolkit. We
reset the decision tree rule using the NPP and NDVI according to the aforemen-
tioned information. The WEKA toolkit is a collection of machine learning algo-
rithms for data mining tasks. It contains tools for data pre-processing, classification,
regression, clustering, association rules and visualization. It is also very suitable for
developing new machine learning schemes.
